Every upload, checked by AI before it goes live
Uploads wait in a queue, get analyzed by three models, and go live if they're safe.
30K+videos / day
Node.jsBullMQRedisGeminiLangfuse
Shivansh Goyal · Full-stack engineer
I build what runs behind the app.
Backend-heavy. Video pipelines, feeds, payouts, and fraud checks for a social video app.
Now: open to full-time backend and full-stack roles
The app I work on, end to endClick any part
01Alerts
394 alerts a week. Almost none mattered.
I built a classifier that groups errors, has an AI agent check the impact, and only posts what's real.
Switch between the two
Errors in0
Incidents0
Slack posts0
02Work
Five systems I built. Try each one.
Software Incubator
Jul 2025 — Aug 2026 · Backend internA different feed for every person
Ranked from what they've watched. New users start with what's popular.
300K+feeds / day
Pick someone to build their feed.
TypeScriptpgvectorVertex AIRedis
Every tap and decision, stored for analysis
Events flow through a queue into ClickHouse. Duplicates are stored once.
75M+events / day
0events since you opened this page
App0
→
Queue0
→
ClickHouse0
ClickHouseBullMQRedis
The payout system
Cash-outs paid through Tremendous, with a ledger that always balances.
84K+payout orders
Request
Risk check
Order
Webhook
Ledger
| Order | Account | Debit | Credit |
|---|---|---|---|
| Empty | |||
StrapiPostgreSQLTremendousHMAC webhooks
Catching fake accounts, not real people
Every signup gets a risk score. Suspicious accounts can verify their ID and face to get back in.
9Maccounts checked
VPN or proxy
Emulator
Many accounts, one device
Face used on another account
Verdict
Real
realsuspiciousfake
Rewardson
ID check
DocumentSelfieFace match
VerisoulGeminiRedisPostgreSQL
Ukudala
Dec 2024 — Jun 2025 · Backend developer45+AWS Lambda functions
10K+live WebSocket connections
<100 msmessage delivery
03Builds
Side projects. Use them right here.
04Book
Hiring? Let's talk.
Pick a time for a 30-minute call.